Background - “Expected” metrics

• In sports analytics, “expected” metrics are used to evaluate player or team performance
  • Soccer: Expected goals (xG) \(^{[1]}\), expected assists, expected points
  • Basketball: Expected points, expected possession value \(^{[2]}\)
  • Football: Expected Yards after Catch \(^{[3]}\)
• Many uses for these metrics
  • Compare a player’s output to an objective baseline
  • Evaluate a team’s performance by eliminating randomness

Expected Yards After Fielding (xYAF)

• Given the current conditions when fielding a punt, how many yards do we expect the punt returner to gain

• Takes into account

  • Punt position
  • Fielding position
  • All defender positions (relative)
  • Punt hang time

• Should return an accurate estimate of how many yards the player gains based on these factors

Data

• NFL 2022 Big Data Bowl
  • “The annual sports analytics contest from NFL Football Operations challenges members of the analytics community to contribute to the NFL’s continuing evolution of the use of advanced analytics.”
  • Last year’s winner: “A Defensive Player Coverage Evaluation Framework”\(^{[3]}\)
  • This year, focuses on Special Teams Plays
  • Contains 19,979 plays from 2018 to 2020
• After filtering, we predict based on ~2000 returned punts

Linear Regression

Linear Regression

Linear Regression

Logistic Regression

Best Subset Selection (Forward, Backward)

Forward Subset Selection

Best Subset Selection (Forward, Backward)

Forward Variable Selection

Best Subset Selection (Forward, Backward)

Forward Variable Selection

Best Subset Selection (Forward, Backward)

Backward

Best Subset Selection (Forward, Backward)

Backward Variable Selection

Best Subset Selection (Forward, Backward)

Backward Variable Selection

Random Forests

Random Forests

Random Forests

LASSO/Ridge Regression

Actual vs Predicted Lasso

GAMs

GAM Model

GAM

Actual vs Predicted

Splines

Actual vs Predicted

Boosting

Actual vs Predicted

Model Comparison

MSEs

Evaluation

Problem: Some methods assume normally distributed data

Evaluation

Possible solution: Variable Transforms

Evaluation

Better, but also doesn’t handle negative values.

Examples

An 85 yard punt return by Diontae Johnson

Examples

A 2 yard punt return by Richie James Jr.

Works Cited

[1] Sam Green. Assessing the performance of Premier League goalscorers. Stats Perform, 2012.

[2] https://grantland.com/features/expected-value-possession-nba-analytics/

[3] https://www.nfl.com/news/next-gen-stats-intro-to-expected-yards-after-catch-0ap3000000983644

[4] Peng et. al., A Defensive Player Coverage Evaluation Framework, https://www.kaggle.com/model284/a-defensive-player-coverage-evaluation-framework